Training capable AI models cheaply on affordable research hardware

Original: Can we post-train LLMs from scratch across Chat, Reasoning, and Code Gen on academic-affordable compute and match industry-released variants?

Source: x.com ↗